See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3869351

Processors for mobile applications

Conference Paper · February 2000

DOI: 10.1109/ICCD.2000.878354 · Source: IEEE Xplore

CITATIONS READS
33 2,730

4 authors, including:

Farinaz Koushanfar J.M. Rabaey

University of California, San Diego University of California, Berkeley
306 PUBLICATIONS   14,183 CITATIONS    526 PUBLICATIONS   33,696 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Privacy-preserving machine learning View project

High-level garbled-circuit framework View project

All content following this page was uploaded by Farinaz Koushanfar on 23 December 2015.

The user has requested enhancement of the downloaded file.

 Processors for Mobile Applications
Farinaz Koushanfar 1, Vandana Prabhu 4, Miodrag Potkonjak 3, Jan M. Rabaey 2
{1 EE Department, 3 CS Department}, University of California, Los Angeles, CA
2
EECS Department, University of California, Berkeley, CA
4
Tensilica, Inc. Santa Clara, CA
Abstract:
Mobile processors form a large and very fast growing segment
of semiconductor market. Although they are used in a great variety
of embedded systems such as personal digital organizers (PDAs),
smart cards, internet appliances, laptops, smart badges, cellular
phones, wearable computers, and sensor networks, they share the
common need for low power, code density, security, cost sensitivity
and multimedia and communication processing.
The goal of this paper is to review the field of processors for
mobile applications. We survey a spectrum of processors, their
system software, and the accompanying hardware components.
The emphasis is on classification and identification of major
technology and architecture trends. Companion to this paper is a
WWW page [Mob00], which provides comprehensive additional
material about mobile processors.
1 Introduction
1.1 Motivation
In 1970, Intel introduced the first single chip microprocessor,
the Intel 4004. It had 2,600 manually placed transistors and
clocking frequency of 100 KHz. Since then, the consistent progress
in semiconductor integrated circuits technology, the introduction of
new architectures, and new synthesis tools has enabled a
spectacular growth in performance. For example, the computational
power of general-purpose processors has increased by more than
50% in the last 15 years. Today's state-of-the-art workstation
microprocessors feature tens of millions of transistors, and run at
GHz clock frequencies. In addition to these quantitative changes,
we have also witnessed a major product diversification.
Microprocessor technology is used for the manufacturing of
microcontrollers, digital signal processors, graphic processors, and
video processors.
In the last several years, the communication processor has
emerged as one of the fastest growing processor markets. Both
network processors for broadband optical communication and a
variety of processors for wireless applications have been attracting
a great deal of research and development, as well as market
interest. Furthermore, the emergence of the next generation of
distributed, ubiquitous computation, as embodied in sensor
networks, ensures the long-term importance of mobile processors.
Our goal in this paper is to survey mobile processors. More
specifically we summarize the main types of processors for mobile
applications, their system software, and the related system
hardware components. We classify mobile processors in four main
groups: computation-oriented (e.g. laptops, handhelds, PDA, and
calculators), communication-oriented (e.g. wireless phones),
media-oriented (e.g. MP-3 players and digital cameras) and devices
for emerging applications (e.g. smart cards and sensor networks).
The emphasis is on identifying major trade-offs and trends. We also
explore the relationship between the mobile processors, technology
and applications.
1.2 What defines a Processor for Mobile Applications
The processors under discussion are used in a wide spectrum of
embedded systems, including laptops, personal digital organizers,
calculators, wearable computers, cellular phones, mobile internet
terminals, MP-3 players, digital cameras, digital cam-coders, smart
cards, and sensor networks nodes. These systems differ
considerably in terms of their computational and communication
requirements, and the types of real-time constraints. Nevertheless
mobile processors share a number of common characteristics.
Most importantly, processors that are used in mobile applications
are subject to design metrics, that emphasize cost, time-to-market,
and low power. In contrast to desktop processor, just-in-time
computing takes precedence over the traditional maximum
performance metric.
Because of the constrained resources of power and cost, and the
real-time computation requirements, the processors for use in
mobile applications posses a number of distinct characteristics.
Examples of such characteristics are:
• Limited Programmability. Almost all mobile systems have a
limited functionality and require a smaller number of
operation modes than their wired counterparts. In many cases,
the application is very well defined, and does not change much
over the life of the mobile product. Therefore, limited
programmability often provides the best trade-off.
• High I/O to Computation Ratio. In many mobile processors,
communication is the primary feature. Even when data
processing is needed, most often it requires relatively simple
computations, at least in comparison with typical desktop
programs.
• Stream data processing. Most mobile applications include
some form of multimedia processing. Initially this was
restricted to voice, but newer devices feature a richer media
content.
1.3 Paper Organization
The reminder of the paper is organized in the following way. In
the next section we survey and categorize mobile processor
according to domains of their applications. After that we discuss
major trends in mobile embedded processors. Next, we cover most
relevant technologies, including power sources. Section 5 presents
an overview of system software for mobile processors. Before
concluding, we outline some of the most exciting (and most likely)
future trends.
2 Mobile Processor Survey
In this section, we classify embedded mobile processors into
four groups according to their targeted applications. Next, we
review the most popular embedded processors used in mobile
applications today, and explain the technical specifications of these
processors.
2.1 Mobile Computers
Historically, mobile processors have been used mostly as
computers. While initially the focus was on calculators, recently it
has shifted to notebooks and PDAs. The emphasis in today’s
laptops is to provide complete workstation functionality, with no
more than 25% reduction in performance. PDAs are on the other
side of the mobile computer spectrum, providing a limited
functionality under strict cost and power constraints.
• Laptops: Laptops usually use a lower power version of
desktop microprocessors. To reduce power consumption, most of
current mobile processors operate at a lower voltage and
frequencies than the corresponding desktop versions. As of
today, the highest clock rate for laptops is 700Mhz compared to 1
GHz. Intel’s low end processors such as Tillamook, Celeron, and
Deschutes operate at 300 to 400 MHz, and consume between 6
and 10 W. The PowerPC 750-400 operates at 440 MHz and
consumes 8 W. This compares to the desktop PIII that operates at
1GHz, consumes 23 W, and is considered a power-efficient
workstation processor. Battery life for a laptop today ranges
between 3 and 6 hours. The market has been dominated for years
by AMD’s and Intel's Pentium mobile processors. Recently,
Transmeta announced the Crusoe [Cme00], a VLIW processor
that uses binary translation (code morphing) to run Intel-
compatible code, while reducing processor complexity and hence
improving power efficiency with little performance reduction.
Unfortunately, technical details of the processor are not available
in the open literature.
• Hand-held PCs and palm PCs: Hand-held PCs deliver PC
functionality in super-portable form, while offering more
computing power than a palm-sized PC. They cannot run full
versions of the application software, and lack internal hard disks.
They use static memory rather than a hard disk for internal
storage. Thereby offering a longer battery life. A crucial design
feature of the Palm PC is the users' ability to input and retrieve
information in a variety of formats. The Palm PC allows users to
quickly access and input information by voice, ink, a soft
keyboard, or handwriting recognition, using natural and
simplified characters. The MIPS RISC processor in one of its
many varieties is the favored processor. For example, Compaq
uses PR31700 in PC Companion, NEC uses VR4102 in Mobile
Pro, Philips, Samsung, and Sharp use PR31700 in Velo,
Infomobile, and Mobilon respectively, and Toshiba uses TX 3900
in Genio.
• Personal Digital Assistants (PDAs): The PDA was
pioneered by Apple Computer, which introduced the Newton
MessagePad in 1993. Shortly thereafter, several other
manufacturers offered similar products. Initially, PDAs had only
modest success in the marketplace. Now, PDAs are common
gadgets. A typical high-end PDA can function as a cellular
phone, fax, and personal organizer. Unlike portable computers,
most PDAs are pen-based, using a stylus rather than a keyboard
for input. Some PDAs can also react to voice input by using
voice recognition technologies. Palm OS is by far the most
popular operating system for PDAs. The primary processor
architecture for both Palm and all its licensees is the Motorola's
Dragonball chip.
2.2 Wireless Communication Devices
• Wireless phones: The wireless (mobile) phone market is one
of the largest and fastest growing high tech markets. In 1996
there were less than 100 million subscribers, in 2000 more than
250 million, and it is expected that by 2008, there will be more
than 1 billion. The wireless phone processor market is currently
dominated by the ARM microcontroller - TI DSP processor
combination. TI has more than 50% and Lucent more than 25%
of this DSP market. Two other larger players are Motorola and
Analog Devices, each with approximately 10% of the market.
The details on the TI DSP processor in mobile phones are
discussed in section 4.
• Pagers: Charles Neergard invented paging in 1949 while he
was in the hospital and needed assistance. Today, more than 20
million people in US have pagers. A recently developed paging
unit, i.e. the 8-bit mobile data receiver, is spurring the growth of
mobile data computer applications. They connect directly to a
computer serial port or plug into a PC Card expansion slot, and
allow any type of data to be sent directly to a roaming computer,
Thus providing applications such as spreadsheet and calendar
updates, faxes, and graphics. Pagers mainly use 8-bit
microcontrollers as their computational engine.
2.3 Media Electronics - Information Appliances
The next step in the Internet revolution, in addition to wireless
Internet, are Internet Appliances. This is one of the fastest evolving
mobile markets. We only survey a few of them. .
• MP3 Players: The MP3 players available on the market
today are positioned as PC companion pieces, rather than
standalone consumer devices. The most recent product is the
Cirrus Logic EP7209, which replaces the three to five ICs used in
current MP3 players. Such devices include a fixed-function DSP,
an external 8- or 16-bit microcontroller, and programmable logic
devices for storage, display and parallel ports. While more than
half of the EP7209's available processing power is devoted to the
decompression of audio algorithms and copy-protection schemes,
the chip, operating at 74 MHz, still can devote 25 MHz to other
unique features.
• Digital Cameras: The worldwide market for digital
cameras is expected to reach $5.4 billion by 2002. Today, digital
cameras provide the major functions needed to preview, capture,
compress, store, transfer and display digital images. A typical
processor for this domain is represented by the Cirrus-IBM-
Polaroid. The three-chip set includes a Sierra Imaging's DSP
ASIC, a SparcLiteRISC processor (MB86831) and an 8-bit
micro-controller (MB89165) from Fujitsu Microelectronics. An
ARM core has recently displaced the SparcLite device as the
central controller.
2.4 Emerging Applications
• Smart Cards: Smart cards are credit card-like devices with
some processing capability. A smart cards main limitation is
memory size - typically 512 bytes of RAM, 16 Kbytes of masked
ROM, and 8 Kbytes of EEPROM. The chip manufactures are
mainly European: Siemens, Philips, Thomson, Motorola UK,
Hitachi, and Oki. All major smart card manufacturers have either
developed or are developing these advanced smart-card chips in
their products (Philips 83C852, Motorola 68HC05SC29, Siemens
SLE44C200, Thomson ST16CF54, European Cascade EP8670,
Crisp 20847).
• Smart Badges: A badge is an identification and access
control system. Badges recently started to evolve and acquired
images, names, bar codes, magnetic stripes, and finally
microprocessors and various types of transponders. The
architecture of an intelligent badge system was first proposed in
Olivetti Research Labs, and was later used by XEROX Parc and
called an active badge. Implementations of badge system to
support context-aware computing, were the Intelligent Badge and
Smart Badge systems, developed at KTH in Stockholm and
Wollongong University in Australia.
3 Major Technological Trends significant, but of a lesser scale. Obviously, low power is a prime
design objective. However, this has to be achieved with moderate
The single most important and most influential semiconductor technology due to high cost sensitivity. It is also interesting to
trend has been a three-decade long sequence of technological follow the historical evolution of the three types of processors by
improvements. This progress enabled very-large scale of comparing the data in Tables 1 and 2. For example, we see that the
integration of a single chip, and the realization of remarkably
technology used by a workstation processor in 1996 is superior to
powerful processors. It is interesting and instructive to take a closer the one used by mobile processors in 2000 in terms of feature size
look at key technological data and trends in embedded mobile and number of metal layers. Furthermore, we see a small reduction
processors. To put things in perspective, Table 1.A shows data for in power for mobile processors. The main change is that size went
state-of-the-art workstation and embedded processors that are up by more than three times. It is apparent that the need posed by
widely used in the mid of 2000. Table 1.B shows the same the more complex applications is answered mainly through
information for three types of processors in 1996. The internal additional parallelism, with only a moderate increase in frequency.
organizations of the tables are identical. An interesting trend can be seen in configurable or
The first column shows the design feature. The next four
programmable CPUs. They form a new class of embedded
columns shows minimal, average, maximal and representative processors that allow designers to customize on-chip resources of
values for the features in the workstation processors. This is processor architecture to their application yet use standardized
followed by the same data for embedded and eventually mobile software and development tools that conform to existing ASIC SoC
processors. Representative processors in Table 1.A are Intel design methodologies. RISC/DSP type configurable and
Pentium PIII, AMD 486DX5, and ARM 710. The source of all this synthesizable cores from Tensilica, ARC Cores and Improv
information is the Microprocessor Report [Mpr00]. For each type Systems’ configurable Superscalar/VLIW cores, belong to this
of processor we used between nine and ten samples. For example, class of processors. Tensilica’s Xtensa processor [Gon00] allows
for workstation processors in 2000, we used Compaq Alpha 21264,
extensibility of the instruction set architecture using Tensilica
AMD Athlon, Intel PIII, MIPS R12000, IBM Power3, PowerPC Instruction Extension (TIE) Compiler with DSP rich coprocessor
7400 (G4), Sun Ultra-2, Sun Ultra-2i, and Hal Sparc64-III. For options [Gon00]. ARC Cores offer an architecture that can accept
mobile processors, the samples consist of: SA-110, ARM710, plug-ins to extend a core’s instruction set, register set and bus
SH7708, 401GF, 403GX, VR4121, M32R/D, CF5202, and V851. configuration. Improv Systems provides a programmable platform
A number of conclusions can be drawn from the two tables. for a multi-processor implementation of a configurable datapath.
Most striking is the large technological advantage of workstation Alternate approaches have been to introduce reconfigurable
processors, if only the sheer performance and integration is processors by including programmable logic elements in the fabric
considered. For example, they are more than 7 times faster and 13
of the processor as shown by Chameleon Systems CS2000 family
times larger at average. They use two more metal layers and have a and the Manta processor from BOPS. Clearly, configurable
3 times larger die size. But, they cost 6 times more, and also processors are giving the flexibility to customize processors for
consume 100 times more power than mobile processors. The networking and mobile applications without increasing the design
differences between embedded and mobile processors are also cycle time.
Processor
Table Workstation processor Embedded processor Mobile processor
1.A ARM
Min Avg Max PentiumPro Min Avg Max AMD2940 Min Avg Max
710
Freq(MHz) 110 214 500 200 25 63 133 50 20 52 200 40
8K/8 32K/ 64K/ 16K/
Cache(I/D) 8K/8K 1K 4K/4K 16K/8K 8K/8K 512/0K 4K 8K
K 32K 64K 16K
0.29
0.4u 0.5u 0.35u 0.35u 0.65u 0.8u 0.7u 0.35u 0.6u 1.0u 0.6u
IC process u
4M 4M 4M 3M 3M 3M 3M 3M 2M 2M 2M
5M
Voltage(V) 3.3 4 5 3.3 3.3 4 5 3.3 2.0/3.3 3.6 5 5
2.3 0.57
# of xtors 6.4M 15M 5.5M 0.35M 1.43M 2.53M 1.2M 0.256M 0.695M 2.1M
M M
Die (mm2) 84 226 335 196 25 100 217 119 25 44 82 25
Power (W) 9 25 40 35 0.5 2.1 4.5 1.7 0.12 0.54 1.1 0.424
mfg cost $25 $150 $375 $175 $8 $25 $75 $20 $4 $10 $18 $9

Processor
Table Workstation processor Embedded processor Mobile processor
AMD486D ARM
1.B Min Avg Max Intel PIII Min Avg Max Min Avg Max
X5 710
Freq(MHz) 296 550 1000 1000 25 125 400 133 25 75 233 25
16K/ 64K/ 512K/ 16K/16 32K/32 16K/
Cache(I/D) 16/16/256 1K 16K 2K 8K 8K
16K 64K 1M K K 16K
0.18u 0.25u 0.29u 0.18u 0.35u 0.7u 0.35u 0.25u 0.8u 0.8u
IC process 0.18u 0.5 3M
6M 5M 4M 6M 3M 3M 3M 3M 2M 2M
Voltage(V) 2.5 3 3.3 2.5 1.35/3.3 3.3 5 3.3 2.0/3.3 5 3.3 3.3
0.341
# of xtors 3.8M 26M 130M 24M 0.35M 2M 28M 1.2M 0.341M 2M 7M
M
Die (mm2) 83 185 477 106 27 62 119 43 22 55 154 40
Power (W) 13 40 75 23 0.4 2.6 10 2.75 0.12 0.29 0.42 0.12
Mfg cost $40 $130 $330 $40 $8 $16 $39 $11 $4 $18 $65 $9

Table 1 State-of-the-art of Different Processors: Table 1.A (1996), Table 1.B (2000)
Power sources are among the most crucial components when Comp. Hyp. 386 Duo 230 Duo 270c Duo 280c Avg
designing mobile embedded systems. Batteries are currently Processor 4% 17% 9% 25% 14%
widely used as the power supply of choice for mobile embedded Hard disk 12% 9% 4% 8% 8%
systems. Portability constrains the size and weight of the batteries. Backlight 157% 25% 26% 25% 23%
In addition, frequent battery replacements and frequent and long Display 4% 4% 17% 10% 9%
Modem N/A 1% 0% 5% 2%
battery recharging are also undesirable. Figure 2 shows a
FPU 1% N/A 3% N/A 2%
simplified view of the growth in the battery technology vs. the Video 26% 8% 10% 6% 13%
microprocessors technology. As can be seen from the chart, Memory 3% 1% 1% 1% 2%
Moore's law is not applicable to batteries. However, as new Other 33% 35% 28% 22% 30%
applications emerges on embedded processors, the power
requirements for mobile processors increase. Figure 3 illustrates Total 6W 5W 4W 8W 6W
Table 4 - Power Consumption in Laptops
the trends in total processor's power consumption as the area and
the frequency of the die increases.

4 Case study – DSP processors

16
14
Processor(MIP
12
10
8 Hard Disk
4 Memory
2 Battery (energy
1
0 1 2 3 4
Time
Figure2 Improvement compared to year 0
Table 4 provides information about where power is spend in
laptops. Backlight and video are two major sources of power
consumption. Hard disks, processors, and displays are also
consuming high portions of the power. Wireless network interface
card (NIC) is also a significant source of power consumption. The
limited power nature of the batteries compounded with their slow
growth, implies a serious need to explicitly design the mobile
embedded system for low power. This system level low power
optimization should include the entire synthesis at all levels
including circuit, logic, software, protocols, and algorithms. The
primary is to optimize the energy metric while meeting QoS
constraints. Recently a number of alternatives were proposed in
order to overcome the limitations of batteries. An example of such
methods was discussed in [Ami97].
Figure 3 Power consumption vs. clock freq. x die area
DSP processors form one of the most important classes of
mobile embedded processors. DSP hardware sales generated over
$3 billion of revenue last year and they have one of the highest
CAGRs (Compound Annual Growth Rate) in the semiconductor
industry. In particular, DSP processors are a mandatory component
in today’s mobile phones.
It is very instructive constructive to analyze the trends in DSPs
in the last two decades. The DSP architecture has been strongly
impacted by both the evolution in technology and the targeted
applications. One of major architectural novelties of the DSP
processor was the special multiply-accumulate (MAC) unit. In
1980, a general-purpose programmable DSP had more than 10
times MAC performance advantage over the microprocessor.
[Ack98]. In 1985, the MAC performance advantage of DSPs over
microprocessors shrank to 5. By 1995, the MAC performance of
DSPs and microprocessors became comparable. DSPs have seen a
slow performance improvement so that today's microprocessors are
outperforming DSPs in MAC dominated applications [Ack98]. On
the other hand, DSPs developed several alternative advantages over
microprocessors. They have significantly lower-power lower-cost
signal processing engines that have MOP/mm2 and MOP/mW
ratings that are an order of magnitude better than microprocessors.
The area and power efficiency of DSP processors is optimized at
several design levels: at the architecture and algorithmic level by
modification of arithmetic modules through optimized memory
organization, by use of special logic families, and through using
more efficient and power-sensitive CAD tools [Ver00].
Advances in digital communications and pervasive computing,
is putting new and even more aggressive demands for low power,
low cost, and programmable DSPs. Examples of the growing
mobile embedded applications include cellular terminals, smart
phones, real-time video communications, 3D graphics and wireless
base-stations. The DSPs in use today are not promising candidates
for the applications mentioned above. A new generation of DSP
processors with support for complex, multi-threaded, real-time
synchronous applications is required. Parallel architectures are
required to meet the desired performance. The new applications
will have much larger memory and bandwidth requirements than
before. The limited time to market frame implies a need for a
compiler-driven programming environment that supports parallel
programming and debugging. Portability requires an even further
reduced power consumption, which is rapidly becoming the most
critical design issue. Hence, New power-saving techniques have to
be introduced. For instance, dynamic voltage scaling is an attractive
option to maintain the performance at worst-case conditions while
minimizing the power under normal conditions.
 Among the latest attempts toward realizing the next generation
DSP core is TI's TMS320C55x core. The C55x DSP core has 2
MAC and 4 ALU units. It has a variable instruction length, from 8
to 48 bits. It runs on frequencies up to 400 Megahertz (MHz),
delivering the raw processing power (up to 900 MIPS and 800
MMACS) needed to bring graphics-intensive Internet applications,
digital still imaging and digital audio, as well as real-time video
conferencing to wireless handsets. The C55x DSP core delivers
ultra-low power consumption of 0.05 milliWatts per MIPS. Also, it
uses automatic power management mode and configurable idle
domains to father reduce power. Additionally, in the next
generation voice-only handsets, the C55x DSP core will enable a
phone's battery life to operate at least a half-day to one week longer
than today's leading C54x DSP-driven devices. In order to
facilitate debugging it provides special emulation features.
5 Beyond Processors - System Software
In this section, we discuss the impact of system software, and in
particular operating systems, on mobile processor markets and
architectural trends. In addition, we briefly mention compilers and
software utility tools for mobile processors.
5.1 RTOS for mobile processors
Operating system is a program that provides an interface
between application programs (and often a user) and the computer
hardware resources. The four main tasks of both general-purpose
computing and embedded systems operating systems are: 1)
process management; 2) interprocess communication and
synchronization; 3) memory management; 4) input/output (I/O)
management [Sil88, Tan97]. In the general-purpose
microprocessor market, the operating system (OS) has been of
great importance. For example, the most popular general-purpose
operating systems are MS-DOS and Windows, and with this comes
the implication that the Intel's x86 architecture is the dominating
hardware platform for both personal computers and workstations.
Similarly, the next two most popular architectures are the PowerPC
due to the Mac OS used in Apple's computers, and RISC processor
that run UNIX (e.g. Sun) used in scientific and engineering
workstations [Com98]. More recently, Linux has been attracting a
high level of interest and has been ported to numerous platforms.
Following this parallel, there is a growing consensus that
operating systems for mobile embedded processors are also of
ultimate importance for market success of processors. While there
are numerous similarities between general purpose OSs and
embedded systems OSs, there are also significant differences. For
example, in embedded systems, the operating system not only
provides the interface between application and hardware, but often
also enables the handling of tasks with real-time requirements.
Real-time systems are systems where proper functionality requires
both functional correctness as well as the correct timing behavior of
the system [Sli99]. The main properties that make a mobile
embedded system different from a general-purpose desktop system
include:
• Hardware constraints in terms of working memory,
nonvolatile storage and power consumption are significantly
stricter;
• User interfaces with significantly limited capabilities
compared to the one found in a general-purpose desktop
systems;
• Often application specific software and hardware is used
rather than the traditional general purpose hardware and
software, employed in desktop systems;
• Streaming media applications (e.g. voice, audio, video) which
requires specific real-time requirements are often the
dominant type of application;
These features of mobile processors have numerous
ramifications. For example, although computational performances
of general-purpose processors highly outperform the performances
of DSP processors, the latter have I/O characteristics that are much
better suited for media applications [Bru96] and therefore often
imply better overall system performance. Furthermore, because
use of mobile embedded systems is not restricted to a fixed point,
security and privacy are of significantly higher importance.
Finally, note that modern mobile processors most often target a soft
real-time operating system (RTOS) rather than a hard RTOS. For
hard RTOS, the most important requirement is that throughput and
computational requirements latencies are deterministic and that all
deadlines are always met. In contrast, for soft RTOS, the main
issue is tailoring the operating system to the strict hardware and
power requirements, while the deadlines for real-time tasks
occasionally may not be met. Comprehensive survey of embedded
and real-time systems trends and issues is given in [Li97].
5.2 Compilers
In addition to the OS, utility programs play an important role in
the acceptance and use of mobile processors. Utility tools are
usually organized in integrated development environments, which
include tools such as project builders, source level debuggers, event
analyzers, performance profilers, run-time error checking tools,
graphical code browsers, specialized text editors, and version
control systems. The most important utility tools for mobile
processors are compilers, linkers and loaders [Lev99].
Development of power and memory utilization sensitive compilers
is still mainly in the research phase.
6 Beyond Processors -Emerging Mobile Applications
6.1 Towards wireless networking
The current generation of mobile applications ranges widely
from palm-sized, stylus-based class of PDA applications such as
personal organizers, (e.g. phone/address book), wireless links to
web applications, (such as email and web browsing) to keyboard-
based desktop type of applications (such as word processing and
spreadsheets). The Global Positioning System (GPS) has improved
the positioning of mobile systems. In all these applications, the
capability to transfer and synchronize data between PDAs and
desktop systems has played an equally important role in the
acceptance of mobile applications. Current portable devices use
infrared links (IrDA, Infra-red Data Association) to communicate
with each other. Although infrared transceivers are inexpensive,
then have a limited range (typically one or two meters), are
sensitive to direction and require direct line-of-sight and can in
principle only be used between two devices.
Radios, on the other hand, have much greater ranges, can
propagate around objects and through various materials and
connect many devices simultaneously. The availability of a low-
cost ad-hoc radio link between mobile computers, mobile phones
and other portable hand-held devices along with connectivity to the
Internet is expected to revolutionize the world of mobile computing
and communications. Bluetooth [Har98] and HomeRF [Hom00]
aim at providing ubiquitous computing by eliminating wired
channels, calling for interoperability between various portable
devices with the flexibility of a wireless solution.
6.2 Sensor Networks
The availability of cheap wireless connectivity is bound to
create exciting new opportunities. Examples of this is distributed,
large-scale sensor networks. This has been the focus of some
research projects such as the PicoNode project [Rab00] and
[Mob00]. These networks require versatile, self-organizing, multi-
functional and adaptive elements for communication and
computation, yet have minimal form factor, energy and cost to
make them feasible for large-scale coverage. This has led to the
idea of a highly energy-efficient singe-chip nodal architecture that
has a sensing device and its conversion into a digital representation
integrated with local computation and communication capabilities.
Support for the network architecture and the accompanying
protocols is also a requirement for the single node. Figure 6 shows
a possible architecture. Flexibility and energy-efficiency are at the
core in the design of the PicoNode processor. It is an emerging
belief that this is best accomplished though the use of
reconfigurable co-processors that use a programming-in-space
approach rather than the programming-in-time approach of the
current mobile processors.
Figure 5 PicoNode’s conceptual architecture
7 Conclusion
Recently, mobile applications and mobile programmable
processors have attracted a great deal of attention. We surveyed
embedded mobile processors, and presented their most important
characteristics, their system software, and related technology. The
emphasis was on classification and identification of major
technology and architecture trends. Companion to this paper is a
WWW page [Mob00] with additional comprehensive information.
View publication stats
8 References
[ack98] Ackland,B ;Nicole, C . High performance DSPs - What's hot and
what's not? , Proceedings of International Symposium on Low Power
Electronics and Design, Monterey, CA, ACM 1998:1-6.
[Ami97] Amirtharajah, R.; Chandrakasan, A.P. Self-powered signal
processing using vibration-based power generation. IEEE Journal of Solid-
State Circuits, May 1998, vol.33, (no.5):687-95.
[Bru96] Brustoloni, J.C.; Steenkiste, P. Effects of buffering semantics on
I/O performance. Second USENIX Symposium on Operating Systems
Design and Implementation (OSDI), Seattle, WA, Oct 1996, vol.30, spec.
issue:277-91.
[Bwr00] http://bwrc.eecs.berkeley.edu
[Com98] Comerford, R. Pocket computers ignite OS battle. IEEE
Spectrum, May 1998, vol.35, (no.5):43-8.
[Cme00] Cmelik; Robert F., Ditzel; David R. Kelly; Edmund J. Hunter;
Colin B., Laird; Douglas A. Wing; Malcolm John , Zyner; Grzegorz B. US
patent 6031992: Combining hardware and software to provide an improved
microprocessor , Feb 29, 2000.
[Gon00] Gonzalez, R.E. Xtensa: A Configurable and Extensible Processor.
IEEE Micro Mar-Apr 2000.
[Har98] Haartsen, J. BLUETOOTH- The universal radio interface for ad
hoc, wireless connectivity. Ericsson Review no.3, 1998.
[Hom00] http://www.homerf.org
[Li97] Yanbing Li; Potkonjak, M.; Wolf, W. Real-time operating systems
for embedded computing. International Conference on Computer Design.
1997: 388-92.
[Lev99] Levine, John R. Linkers and loaders. San Francisco : Morgan
Kaufmann, 1999.
[Luc00]http://www.lucent.com/wirelessnet/customers/seminar98/erickson/s
ld002.htm
[Man99] Mann, S. The Wireless Application Protocol. Dr. Dobb's Journal,
Oct. 1999, vol.24, (no.10):56-66.
[Mob00] http://www.cs.ucla.edu/~farinaz/EE233.htm
[Mpr00] http://www.mpronline.com
[Pas00] Passani, L. Creating WAP services. Dr. Dobb's Journal, July 2000,
vol.25, (no.7):70-9.
[Pat99} Patiyoot, D.; Shepherd, S.J. Cryptographic security techniques for
wireless networks. Operating Systems Review, April 1999, vol.33,
(no.2):36-50.
[Rab00] Rabaey, J.M. ; Ammer, M.A.; da Silva, J.L.; Patel, D.; Roundy, S.
PicoRadio Supports Ad Hoc Ultra-Low Power Wireless Networking. IEEE
Computer Magazine, July 2000:42-8.
[Sil88] Silberschatz,A ; Peterson,J.L. Operating systems concepts. Alternate
ed. Reading, Mass. : Addison-Wesley, c1988.
[Sli99] Slijepcevic, S. Review of Embedded Operating Systems - Windows
CE and EPOC. UCLA Class report, Dec 1999:
http://pixies.cs.ucla.edu:8080/home/review.doc
[Sub98] Subramanyam, V.; Joshi, A. Security in mobile systems. IEEE
Symposium on Reliable Distributed Systems, 1998: 407-412.
[Tan97] Tanenbaum, A.S. Woodhull, A. Operating systems : design and
implementation. 2nd ed. Upper Saddle River, NJ : Prentice Hall, 1997.
[Ver00] Verbauwhede, I; Nicol, C. "DSP Processors for Wireless
Commmunications", Invited tutorial, ISSCC, Feb 2000.
[Win99] Winton, G. Extending the PalmOS. Dr. Dobb's Journal, Nov.
1999, vol.24, (no.11):50-5.
[Wol99] Wolk, R.H. Fuel cells for homes and hospitals. IEEE Spectrum,
May 1999, vol.36, (no.5):45-52.